# Perfusion using ASL-MRI

## Team

Moss Zhao, Flora Kennedy McConnell, Yuriko Suzuki, Logan Zhang

### Collaborators

Thomas Okell, Peter Jezzard (FMRIB Centre)

Bradley MacInthosh (Toronto)

Manus Donahue (Vanderbilt)

Marco Casterllaro, Alessandra Bertoldo (Padova)

Illaria Boscolo Galazzo (Verona)

David Thomas, Xavier Golay (UCL)

Esben Petersen (Utrecht)

Matthias van Osch (Leiden)

Matthias Gunther (Bremen)

## Description

Arterial Spin Labelling (ASL) is an MRI method for imaging the delivery of blood to body tissues, and is most widely used in the brain. By manipulating the the magnetization of blood water before entering the tissue - by radio frequency inversion - the blood is labelled without the need to introduce a contrast agent. This labelled blood can be seen in an image of the tissue along with the contribution of the (static) tissue itself. Upon subtraction of a control image of the tissue, without any labelling having been performed, the delivery of blood alone can be measured.

From ASL images the tissue perfusion can be quantified using a kinetic model of blood flow. Often ASL measurements are made at a single time point post-labelling, which leaves the quantified perfusion values subject to some uncertainty due to variations in the time taken for blood to pass form labeling region to the tissue or dispersion effects of the label during its passage. One way to improve the quantification and also extract more information about the vasculature is to make a series of measurements at various post-labelling times and fit a kinetic model to determine a number of parameters.

### Inference

For ASL inference goes beyond conventional model fitting (parameter estimation) to include both prior information about the model parameters as well as producing estimates of the uncertainty of the estimated parameter values. For example, we know that the arrival of labelled blood water in a tissue region is reasonably constrained since it is limited by the distance the blood has to travel from labeling to imaging regions. This can be encoded into the analysis by means of a prior probability distribution for the parameter in a Bayesian inference method. The key advantage being that this does not impose 'hard limits' on the parameter, so if the data was really generated from an arrival time outside the 'normal' range, as would be the case in brain imaging in stroke, this is still permitted - the data can override the prior.

We use a form of Bayesian inference based a Variational Bayes approach to solving the generic problem of fitting a non-linear model to data. This algorithm matches closely to classic non-linear least squares (Gauss Newton) but includes 'probabilistic' terms. A great advantage of this scheme is that its computational performance is similarly fast as least squares analysis. We have investigated the use of various types of prior information in combination with more complete models the ASL signal arising from labelled blood in the vasculature.

### Modelling

The simplest ASL kinetic model includes a single compartment into which labelled blood water is delivered and accumulates. This models the rapid exchange of labelled water from the blood into the tissue. More advanced models would include the contribution from blood that has not yet reached the capillary bed - a potentially significant source of contamination of the perfusion signal in the vicinity of large arteries.

In the brain the ASL signal will arise from blood supply to both grey and white matter in the same imaging region, due to the resolution of the technique. These two components have very different kinetics - both perfusion rates and arrival times. This information along with separately obtained estimates of the partial fractions of the two tissues and spatial homogeneity of perfusion (incorporated via spatial priors) can be used to help extract the perfusion in each tissue type.

### Software

Software for the analysis of ASL data to quantify perfusion using the inference techniques developed in the group is available in the BASIL tool as part of the FMRIB Software Library.

## Selected Publications

**Chappell, M.A.**, Groves, A., Whitcher, B., & Woolrich, M.**Variational Bayesian Inference for a Nonlinear Forward Model.**

IEEE Transactions on Signal Processing, 57(1), 223–236.

**Chappell, M. A.**, MacIntosh, B. J., Donahue, M. J., Guenther, M., Jezzard, P., & Woolrich, M. W.*Separation of Macrovascular Signal in Multi-inversion Time Arterial Spin Labelling MRI.*

Magnetic Resonance in Medicine, 63(5), 1357–1365. doi:10.1002/mrm.22320

**Chappell, M. A.**, Groves, A. R., Macintosh, B. J., Donahue, M. J., Jezzard, P., & Woolrich, M. W.*Partial volume correction of multiple inversion time arterial spin labeling MRI data.*

Magnetic Resonance in Medicine, 65(4), 1173–1183. doi:10.1002/mrm.22641

**Chappell, M. A.**, Woolrich, M. W., Kazan, S., Jezzard, P., Payne, S. J., & MacIntosh, B. J.*Modeling dispersion in arterial spin labeling: Validation using dynamic angiographic measurements.*

Magnetic resonance in medicine, doi:10.1002/mrm.24260

**Chappell, M. A.**, Woolrich, M. W., Petersen, E. T., Golay, X., & Payne, S. J.*Comparing model-based and model-free analysis methods for QUASAR arterial spin labeling perfusion quantification.*

Magnetic resonance in medicine, doi:10.1002/mrm.24372